Quality of Analytics and Runtime Explainability for End-to-End Machine Learning in Edge-Cloud Continuum

Activity: Talk or presentation typesAcademic keynote or plenary lecture


Modern software processing real-time IoT data in Edge-Cloud continuum increasingly utilizes machine learning (ML) features as a part of the software. Such ML features encapsulate advanced ML capabilities and require different integration and management techniques. Based on ML models, these features are powerful but exhibit several contractual problems for service providers and consumers in terms of operational deployment performance, quality of inferences, data quality effects and explainability, to name just a few. Achieving key objectives, like faster serving time, cheaper operation cost, and higher reliability of inferences, is of paramount importance in developing and operating such AI/ML software in Edge-Cloud continuum. This calls for new research to address concerns about quality of analytics (QoA) and its impact on ML services and end-to-end ML serving. In this talk, we will present the QoA4ML framework for qualifying ML services and runtime explainability for end-to-end ML serving. We identify and specify contractual concerns covering different aspects of data, ML models and services. Based on that we develop policies for ML service contracts, providing runtime explainability and adaptation for ML serving. We will discuss our work with realistic applications of object detection, malware detection, and predictive maintenance in the Edge-Cloud continuum.
Period3 May 2024
Event titleInternational Conference on Artificial Intelligence for IoT
Event typeConference
Conference number3
LocationVellore, IndiaShow on map
Degree of RecognitionInternational